Study On Theories And Methods Of Intelligent Multi-criteria Oil Synthesis Analysis And Fault Diagnosis

As an important means for condition monitoring and fault diagnosis of mechanical equipments lubrication system, oil analysis is high efficient and strongly pertinent especially for equipments wear andoil quality . With increas-ing and deep application about fields, the present difficulty also hotspot consists in the feature extraction and knowledge mining from the oil core data like spec-trography and ferrography data. To monitoring system in practice at present, there has been disregard for the impact of noise (a data disturbance) resulted from sampling and analyzing on the feature extract, quite strong subjectivity in the choice of monitoring means and its attributes and excessively depending on experts in diagnosis. These inevitably often result in accuracy, pertinence and universality not enough in oil analysis and diagnosis, which seriously block the effective popularization of oil monitoring technique. As a result based on the characteristics of oil monitoring means and data, the key features about ma-chines wear and oil quality are lucubrated from different points of view in the thesis. Finally, oil monitoring multi-technique and multi– attributes synthesis analysis and diagnosis expert system is set up quietly complete and ability to monitoring and controlling in the course of oil data variety and degree of accu-racy in faulty diagnosis are improved in practice.According to the research of the thesis, a quite whole set of analysis and diagnosis criterions about wear and oil quality is provided for large automatism press line in Shanghai General Motors Company. And the machines'safe reli-ability is guaranteed and considerable effect and economic benefit are obtained.More important, breakthrough has been obtained in perfecting traditional theory and method and progresses have made in exploring new theories and methods. In the basis of the characteristics oil monitoring means and their data, the analytic hierarchy process (AHP), singular value decomposition (SVD), wave packet analysis, K-means clustering and Rough sets are applied to these data to extract feature and mine knowledge, which enrich and develop new analysis processing technique for oil data.In this research, through further analysis the relativity between oil detection information and equipments wear or oil quality, analytic hierarchy process (AHP) is applied to make oil analysis weight model about them to estimation matrix in-cluding every experts'favorite information. The weigh total order of monitoring means and their attributes, which reflects equipment wear, lubrication and oil quality, was solved applying singular value decomposition (SVD) using left and right eigenvector including such informantion. So, the choice of characteristic attributes was also done in oil synthesis analysis so that oil analysis is simplified and posses better pertinence and problem-solving ability, which makes prepara-tion for oil analysis diagnosis criterion. Application of the way to the large press line testifies its validity.The lack of original'three lines method'in the foundation of diagnosis rules about wear state is point out and testified,and multi-line method namely im-provement of'three lines method', is advanced. The method is that after wave packet is used to denoise original monitoring signal, wave decomposition and K-means clustering are applied to de-noised signal to boundary and centroid features so as to build diagnosis rules. The application contrast gives promi-nence to the deficiency of traditional'three lines method'and better efficiency of in analyzing the data. In addition, the necesssity of denoising is proved the three aspects namely boundary and clustering features and relativity between at-tributes.Based on the priority weight(obtained by AHP) of analytical fer-roghraphy's attributes in judging wear and oil quality, application of Rough sets in imperfective information and decision table and tribology character of ma-chine, analytical ferroghraphy knowledge mining model is found to simplify these attributes in wear and oil quality decision system and get detailed diagno-sis rules. Finally, due to objective need for lubrication system condition monitoring excellence and disadvantages of oil monitoring means, function their combina-tion and correlation between monitoring data are analyzed in detailed. In the ba-sis of the above and intelligent reasoning integrating case-based on and rule-based on reasoning, oil monitoring multi-technique and multi– attributes synthesis analysis and diagnosis architecture and expert system merging deep and shallow knowledge, were developed. After continuous feedback and modi-fication through the application in SGM and Shanghai Waigaoqiao shipbuild-ing,the expert system possesses quite good capability to analysis and diagnosis with the merits such as briefness to operate in personal computer, concurrency and long-distance operation in network, and predominance towards online oil analysis.